ICU sedation, delirium & recovery
Early prediction of delirium upon intensive care unit admission: Model development, validation, and deployment
Authors' conclusion
Our early prediction models based on data obtained upon ICU admission could achieve good performance in predicting delirium occurred within 48 h after ICU admission. Our 24-h models can improve delirium prediction for patients discharged >1 day after ICU admission.
Study context
- Population
- Critically ill adults, single university teaching hospital, Taipei, Taiwan, August 2020 to August 2021
- Participants
- 6,238 adults
- Exposure
- Logistic regression (LR) vs gradient boosted trees (GBT) vs deep learning (DL) models predicting delirium at intensive care unit (ICU) admission and at 24 h
- Outcome
- Model performance for predicting delirium
Methodological detailssingle-centre, model development and validation, temporal split training/testing
- Single-centre
- Model development and validation
- Temporal split training/testing
Main finding
As reported in the abstract
| Group | AUROC (highest of admission (ADM) models) | Area under the precision-recall curve (AUPRC) (highest of ADM models) | Brier score | AUROC (highest of 24H models) | AUPRC (highest of 24H models) |
|---|---|---|---|---|---|
| ADM GBT | 0.858, 95% CI 0.835-0.879 | 0.814, 95% CI 0.780-0.844 | 0.140 | ||
| ADM LR | 0.149 | ||||
| ADM DL | 0.145 | ||||
| 24H DL | 0.931, 95% CI 0.911-0.949 | ||||
| 24H LR | 0.842, 95% CI 0.792-0.886 |
AUROC (highest of admission (ADM) models)
ADM GBT0.858, 95% CI 0.835-0.879
ADM LR— not stated in the abstract
ADM DL— not stated in the abstract
24H DL— not stated in the abstract
24H LR— not stated in the abstract
Area under the precision-recall curve (AUPRC) (highest of ADM models)
ADM GBT0.814, 95% CI 0.780-0.844
ADM LR— not stated in the abstract
ADM DL— not stated in the abstract
24H DL— not stated in the abstract
24H LR— not stated in the abstract
Brier score
ADM GBT0.140
ADM LR0.149
ADM DL0.145
24H DL— not stated in the abstract
24H LR— not stated in the abstract
AUROC (highest of 24H models)
ADM GBT— not stated in the abstract
ADM LR— not stated in the abstract
ADM DL— not stated in the abstract
24H DL0.931, 95% CI 0.911-0.949
24H LR— not stated in the abstract
AUPRC (highest of 24H models)
ADM GBT— not stated in the abstract
ADM LR— not stated in the abstract
ADM DL— not stated in the abstract
24H DL— not stated in the abstract
24H LR0.842, 95% CI 0.792-0.886
Observational data: associations, not effects.
Further findings
Incidence of ICU delirium, ADM testing dataset
- Within 24 h
- 32.9%
- Within 48 h
- 36.2%
Features of ADM models
- Features: age, BMI, dementia history, postoperative intensive monitoring, elective surgery, pre-ICU hospital stay, GCS score and initial respiratory rate at ICU admission
- 8
Parts of this Brief are withheld: the automatic checks could not confirm them against the abstract.
Original publication
Early prediction of delirium upon intensive care unit admission: Model development, validation, and deployment. J Clin Anesth. Published 12 April 2023. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →